COVID-19, SARS, and MERS Augmented X-Ray Images

Dataset for the paper entitled "A Large Margin Piecewise Linear Classifier with Fusion of Deep Features in the Diagnosis of COVID-19 ". The dataset was balanced by image augmentation to reach 6179 images. Rotation and translation operations are used to increase the number of images (upsampling) in order to balance the class instances with augmentation.  Images were rotated 5, 10, 15, and 30 degrees, and also instances were translated in a horizontal or vertical direction by 5%,10%, 15%, in SARS and MERS classes which have a few numbers of instances, 134 and 144 images, respectively. Finally, 1072 and 1152 images have been achieved in SARS and MERS classes, respectively.  In COVID-19 class with 423 instances, 15-degree rotation and 10% translation were performed to reach 1269 images in this class.  No augmentation was applied on Normal and typical Viral Pneumonia with 1341 and 1345 images, respectively. All augmented x-ray images can be found in "AugmentedCovidDataset" folder. Five-fold cross-validation (CV) is utilized whereas, 80% of the original labeled data are used as the training set and the remaining instances (20%) are employed as the test set to evaluate the model (unseen instances). The final confusion matrix is computed using the average of the metrics in these folds. In this study, two experiments were carried out for two different classification problems:
Experiment1: Normal, COVID-19, and typical Viral pneumonia
Experiment2: COVID-19, SARS, and MERS pneumonia

The details of the training and test instances in each fold can be found in "5-FoldCV_Experiment1" and "5-FoldCV_Experiment2".